Efficient network disruption under imperfect information: The sharpening effect of network reconstruction with no prior knowledge

Abstract

The problem of network disruption has attracted widespread interests, for it currently appears in a myriad of contexts. However, existing methods require assuming that complete information of networks is known and spurious noise is not included, which is not always available in realistic situations. To overcome these limitations, this paper focuses on the problem of network disintegration with imperfect link information. We present a novel network disruption framework utilizing network reconstruction that is independent of prior knowledge and contains spurious links. Experiments in both synthetic and real networks reveal that, our approach which uses network reconstruction in advance, can obviously improve the network disruption performance. Moreover, we are surprised to find that, the disruption effect of our method is even superior to that achieved by complete information when the noise is relatively small. We name this phenomenon as the “sharpening effect” of network reconstruction, for its feature on reshaping network is similar to the image sharpening. Our results can be very useful for network disruption under incomplete information.